Load balancing in parallel computers: theory and practice by Cheng-Zhong Xu, Francis C.M. Lau

By Cheng-Zhong Xu, Francis C.M. Lau

Load Balancing in Parallel pcs: idea and perform is ready the basic software program means of load balancing in dispensed reminiscence message-passing parallel pcs, often known as multicomputers. each one processor has its personal deal with area and has to speak with different processors via message passing. typically, a direct, point-to-point interconnection community is used for the communications. Many advertisement parallel pcs are of this classification, together with the Intel Paragon, the considering computing device CM-5, and the IBM SP2. Load Balancing in Parallel desktops: concept and perform provides a accomplished remedy of the topic utilizing rigorous mathematical analyses and sensible implementations. the focal point is on nearest-neighbor load balancing equipment within which each processor at each step is specific to balancing its workload with its direct neighbours merely. Nearest-neighbor equipment are iterative in nature simply because an international balanced kingdom could be reached via processors' successive neighborhood operations. considering nearest-neighbor equipment have a rather cozy requirement for the unfold of neighborhood load details around the method, they're versatile when it comes to permitting one to manage the balancing caliber, powerful for keeping communique locality, and will be simply scaled in parallel pcs with a right away communique community. Load Balancing in Parallel desktops: idea and perform serves as an outstanding reference resource and will be used as a textual content for complex classes at the topic.

Show description

Read or Download Load balancing in parallel computers: theory and practice PDF

Best design & architecture books

Operational Amplifiers: Theory and Design

Operational Amplifiers – thought and layout, moment variation offers a scientific circuit layout of operational amplifiers. Containing cutting-edge fabric in addition to the necessities, the publication is written to entice either the circuit clothier and the process clothier. it truly is proven that the topology of all operational amplifiers could be divided into 9 major total configurations.

Computer and Information Security Handbook

The second edition of this entire instruction manual of laptop and knowledge security provides the main whole view of laptop safeguard and privateness on hand. It deals in-depth assurance of defense conception, know-how, and perform as they relate to confirmed applied sciences in addition to contemporary advances.

Languages, Design Methods, and Tools for Electronic System Design: Selected Contributions from FDL 2015

This e-book brings jointly a variety of the easiest papers from the eighteenth variation of the discussion board on specification and layout Languages convention (FDL), which happened on September 14-16, 2015, in Barcelona, Spain. FDL is a well-established foreign discussion board dedicated to dissemination of study effects, useful stories and new principles within the software of specification, layout and verification languages to the layout, modeling and verification of built-in circuits, complicated hardware/software embedded platforms, and mixed-technology platforms.

Extra info for Load balancing in parallel computers: theory and practice

Sample text

Number of processes generated at run-time) for the algorithm to achieve optimal or near-optimal load balancing with very high probability. Liiling et al. addressed the issue of whether a process should be migrated only through direct neighbors or not [132]. Algorithms that always pick a random processor among the direct neighbors are called local random algorithms; algorithms that can pick among remote processors are called global random algorithms. They compared local random and global random algorithms through simulation and showed the superiority of global random algo~ rithms.

Oforl<_i,j <_N. 2. Convergence Analysis 41 Proof. (1) Suppose 0 < A < 1. By definition, Ee is nonnegative and doubly stochastic for all c, 1 < c G n. It is easy to show that their product (of multiplication ) preserves the same properties; that is, E is nonnegative and doubly stochastic [14]. (2) Suppose 0 < A < 1. T h e n ( E ) / , / > 0because (Ee)i,i > 0 for all 1 < c < ~. Hence, E N-1 > 0; and E is primitive for 0 < A < 1. yze the convergence of {E t } using the theory of nonnegative matrices and finite Markov chain [14].

To avoid "horizon effect" (water stagnating on a flat surface), the algorithm imposes a minimum distance a process is required to travel. It also imposes a maximum distance a process is allowed to travel in order to cut down on the traveling cost. If we allow these parameters to be tunable at runtime, the algorithm becomes Adaptive Contracting Within a Neighborhood (ACWN for short). Kale compared ACWN and GM as applied to parallel computations of divide-and-conquer and fibonacci programs through simulation [102], and concluded that ACWN performs better than GM in most cases because of the agility of ACWN in spreading workloads around during load balancing.

Download PDF sample

Rated 4.21 of 5 – based on 23 votes